Re: [9fans] GCC3.0 [Was; Webbrowser]

[David Presotto <presotto@closedmind.org>]

[-- Attachment #1: Type: text/plain, Size: 225 bytes --]

I looked at at the benchmarks.  There are lots of times() calls in
inner loops that are very low cost in Unix but pretty expensive in
Plan 9.  Why not do a run under iostats and post the results.  It
might be enlightening.

[-- Attachment #2: Type: message/rfc822, Size: 5014 bytes --]

From: andrey mirtchovski <mirtchov@cpsc.ucalgary.ca>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] GCC3.0 [Was; Webbrowser]
Date: Wed, 5 Feb 2003 21:16:59 -0700 (MST)
Message-ID: <Pine.LNX.4.44.0302052038050.7087-100000@csl>

I've actually used gcc to compile something -- a simple cpu and memory
benchmark called 'ubench'.

I strongly desire such that the 'my OS is faster' flamewars continue on this
list, that's why i'm posting a URL to the ubench source and compilation
instructions for Plan9 (using the GCC3.0 port).

Unfortunately I couldn't be bothered rewriting the benchmark for the native
P9 compiler, so 'my compiler is faster' flame warriors will be left
dissatisfied.

The URL for ubench is:
	http://www.phystech.com/download/ubench.html

to compile under plan9:
	% gunzip < ubench-0.32.tar.gz | tar xv
	% cd ubench-0.32
	% # edit ubench.c and comment out syslog.h on line 25
	% gnu/gsh
	$ gcc -o ubench signals.c cpubench.c membench.c ubench.c
	(alternatively, to compile with optimizations, do:)
	$ gcc -o ubench signals.c cpubench.c membench.c ubench.c -O2
	$ ^D
	% strip ubench
	% ubench

here are the highly scientific results I got:

Celeron 900mhz (average ~40,000 in FreeBSD 4.7 w/ gcc2.95), the machine is a
standalone 9pcdisk/kfs terminal:

no optimizations:

% ubench
Unix Benchmark Utility v.0.3
Copyright (C) July, 1999 PhysTech, Inc.
Author: Sergei Viznyuk <sv@phystech.com>
http://www.phystech.com/download/ubench.html
Plan9 1 0 generic pcdisk
Ubench CPU:    23072
Ubench MEM:    14221
--------------------
Ubench AVG:    18646
%

with -O2:

% ubench
Unix Benchmark Utility v.0.3
Copyright (C) July, 1999 PhysTech, Inc.
Author: Sergei Viznyuk <sv@phystech.com>
http://www.phystech.com/download/ubench.html
Plan9 1 0 generic pcdisk
Ubench CPU:    24992
Ubench MEM:    18097
--------------------
Ubench AVG:    21544
%




Pentium 4, 2Ghz (average ~60,000 on FreeBSD 4.7 w/ gcc2.95, now the machine
has FBSD 5.0 w/ gcc3.2 on it, but is booted in p9 so I couldn't test :), the
machine is running as an auth/cpu/kfs server:

cpu% ubench
Unix Benchmark Utility v.0.3
Copyright (C) July, 1999 PhysTech, Inc.
Author: Sergei Viznyuk <sv@phystech.com>
http://www.phystech.com/download/ubench.html
Plan9 1 0 generic pcauth
Ubench CPU:    39110
Ubench MEM:    33350
--------------------
Ubench AVG:    36230
cpu%


with -O2:

cpu% ubench
Unix Benchmark Utility v.0.3
Copyright (C) July, 1999 PhysTech, Inc.
Author: Sergei Viznyuk <sv@phystech.com>
http://www.phystech.com/download/ubench.html
Plan9 1 0 generic pcauth
Ubench CPU:    43528
Ubench MEM:    44170
--------------------
Ubench AVG:    43849
cpu%

And something else: I just checked how ubench is compiled on freebsd (what
optimizations are used) and gave the same arguments to plan9's gcc. The
result is:


$ gcc -o ubench signals.c cpubench.c membench.c ubench.c -O2 -Wall '-malign-loops=2' '-malign-jumps=2' '-malign-functions=2' -fomit-frame-pointer -s
cpu% strip ubench
cpu% ubench
Unix Benchmark Utility v.0.3
Copyright (C) July, 1999 PhysTech, Inc.
Author: Sergei Viznyuk <sv@phystech.com>
http://www.phystech.com/download/ubench.html
Plan9 1 0 generic pcauth
Ubench CPU:    43863
Ubench MEM:    44170
--------------------
Ubench AVG:    44016
cpu%


as with everything, those results should not be taken too seriously :)

andrey